A microgrid short-term load prediction method based on deep learning

Abstract

The invention relates to a microgrid short-term load prediction method based on deep learning. Historical data is accurately processed and classified through a CNN deep learning method on the basis ofa CNN and an LSTM network, an LSTM network group is constructed through the LSTM deep learning method, and finally, the two deep learning methods are combined through a selector to obtain a CNN-LSTMmodel; and based on the CNN-LSTM mode, a load prediction model based on the deep learning is constructed, the model not only can predict the power consumption load in the microgrid load with higher precision, but also can predict the distributed photovoltaic power generation load in the microgrid load with higher precision, and finally, the prediction precision of the overall load of the microgridis improved.

Description

technical field [0001] The invention relates to a method for short-term load forecasting of a microgrid, in particular to a load forecasting method based on a deep learning network. Background technique [0002] The microgrid is an important way for the grid to absorb distributed new energy, and it is of great significance to the prediction of the internal load and the overall load of the microgrid. At present, the algorithms used for load forecasting include traditional algorithms and modern algorithms. [0003] Due to the existence of distributed new energy with high penetration rate in the microgrid, its load changes and fluctuations are increasing, and the prediction difficulty is greater than that of traditional large grid loads. The use of traditional widely used load forecasting algorithms such as BP neural network and support vector machine and other shallow learning methods can no longer meet the load forecasting of microgrids with huge fluctuations. When the amou...

AI Technical Summary

Problems solved by technology

The use of traditional widely used load forecasting algorithms such as BP neural network and support vector machine and other shallow learning methods can no longer meet the needs of microgrid load forecasting with huge fluctuations.

When the amount of input data is very large, the shallow learning algorithm network itself is a fully connected network, which will lead to too many weights of the entire network, and the performance will drop sharply. Finally, it will limit the maximum number of neurons that can be accommodated in each layer, and then Limit its learning depth and can only stay in shallow learning

Shallow learning uses the gradient transmission method to carry out the error reverse transmission process, which limits its learning depth. When the learning depth increases, the error of reverse transmission will decrease sharply with the increase of depth, resulting in the phenomenon of "gradient disappearance". The transmission effect will only stay in the shallow layer, and basically has no effect on the weight update of the deep layer. The training effect tends to be zero, and even the parameters of the front layer near the input layer cannot be optimized and tend to be random.

Images